1132 Q. Tan et al. / Computers and Mathematics with Applications 61 (2011) 1129–1144
Let J be the set of all the standard parameter systems. It is obvious that there exists a one-to-one correspondence between
J and
˜
X
n
. So the equivalent relation ‘‘≈’’ can be transplanted into J. Let X(T ) denote the equivalent standard form which has
the standard parameter system T .
Theorem 4 ([8]). X
n
=
σ ∈S
n
˜
˜
T ∈J/≈
S∈
˜
˜
T
{X(S)
σ
}.
Denote C(
˜
˜
T ) = {S : S is a parameter system that has the same diagram as T , but its inequalities may not be necessarily strictly
}. X = X(T ), C(
˜
˜
X) = {X(S) : S ∈ C (
˜
˜
T )} is the set of fuzzy equivalent matrices corresponding to C(
˜
˜
T ).
Theorem 5 ([8]). X
n
=
σ ∈S
n
˜
˜
T ∈J/≈
S∈C(
˜
˜
T )
{X(S)
σ
}.
Definition 2 ([8]). The distance between fuzzy similar matrix A and B is defined as:
d(A, B) =
2
n−1
−
i=1
n
−
j=i+1
(a
ij
− b
ij
)
2
A, B ∈ Y
n
. (3)
Theorem 6 ([8]). For any R ∈ Y
n
, there exists R
#
∈ X
n
s.t.
d(R
#
, R) = inf
X∈X
n
d(X, R). (4)
The fuzzy equivalent matrix that is closest to the given similar matrix is called a globally optimal fuzzy equivalent matrix.
Obviously, it has the least lack of fidelity brought about by clustering according to the globally optimal fuzzy equivalent
matrix.
Corollary 1 ([8]). Let R ∈ Y
n
and R
#
be the optimal equivalent matrix of R. Then R ∈ X
n
⇔ R
#
= R ⇔ d(R, R
#
) = 0.
Corollary 2 ([8]). ∀R ∈ Y
n
, let R
#
be the optimal fuzzy equivalent matrix of R and R
∗
be the transitive closure of R. Then
d(R, R
#
) ≤ d(R, R
∗
).
Corollary 3 ([8]). ∀R ∈ Y
n
, let R
#
be the optimal fuzzy equivalent matrix of R and R
∗
be the transitive closure of R. Then
R ∈ X
n
⇔ R
∗
= R
#
= R.
Theorem 7 ([8]). Let R ∈ Y
n
. If there exists R
#
∈ C (
˜
˜
X) ⊆ X
n
s.t. d(R, R
#
) = inf
X∈C (
˜
˜
X)
d(X, R), then t
i
=
b
i1
+b
i2
+···+b
im
i
m
i
(i = 1,
2, . . . , n − 1), where t
1
, t
2
, . . . , t
n−1
are parameters in the parameter system of R
#
and b
i1
, b
i2
, . . . , b
im
i
are elements in R
corresponding to t
i
(i = 1, 2, . . . , n − 1) in R
#
.
Based on the above result, a FCMBP method for calculating the globally optimal fuzzy equivalent matrix could be obtained
as follows [8].
Step 1. Construct a storage of all classes of similar equivalent standard forms,
˜
X
n
/≈ and their parameter systems, J/≈.
Step 2. Let R ∈ Y
n
.
Step 3. Let σ ∈ S
n
.
Step 4. Let
˜
˜
X ∈
˜
X
n
/≈.
Step 5. Calculate R
σ
.
Step 6. Find b
i1
, b
i2
, . . . , b
im
i
(i = 1, 2, . . . , n − 1) in R
σ
corresponding to t
i
in
˜
˜
X.
Step 7. Calculate t
′
i
=
b
i1
+b
i2
+···+b
im
i
m
i
, i = 1, 2, . . . , n − 1.
Step 8. Check whether t
′
i
(i = 1, 2, . . . , n − 1) satisfies the inequalities given by
˜
˜
X. If not, then go to Step 4. If they are, then
go to the next step.
Step 9. Construct a matrix X
′
by using t
′
i
(i = 1, 2, . . . , n − 1) such that X
′
∈
˜
˜
X and calculate d(X
′
, R
σ
).
Step 10. Repeat Steps 4 through 9 until
˜
˜
X runs over all classes of similar equivalent standard forms. And find X
σ
in all X
′
such
that d(X
σ
, R
σ
) is the smallest in all d(X
′
, R
σ
). Then go to Step 3.
Step 11. Repeat Steps 3 through 10 until σ runs over all elements in S
n
. And find X
#
in all X
σ
(σ ∈ S
n
) and σ
#
∈ S
n
such that
d(X
#
, R
σ
#
) = inf
σ ∈S
n
d(X
σ
, R
σ
).
Step 12. Calculate R
#
= X
#
(σ
#
)
−1
. Then d(R
#
, R) = inf
X∈X
n
d(X, R).
The above FCMBP clustering method is an effective method based on the resolution structure, which contains a double
loop procedure. In the external loop, σ runs over all elements in S
n
. For a certain σ ∈ S
n
, Steps 4–9 constitute the inner
loop, in which
˜
˜
X runs over all classes of similar equivalent standard forms in
˜
X
n
/≈. After running over all the combination